Re: [Bacula-users] Getting err=22 message in dbcheck

[Marco v. W. <mv...@pl...>]

Steven A. Falco <safalco <at> optonline.net> writes:

> 
> 
> In other words, the call to rwl_writelock_p is returning EINVAL
> because the rwl is not valid.  (I added a printf right before
> the return statement, so I am sure that this is the place the
> EINVAL (err 22) comes from.)
> 
> Again, dbcheck puts out good data; it just has problems cleaning
> up at the end.  And everything else works - I can run backups,
> do restores, etc.  It is just the catalog backup that has problems.
> 
> I'm not really sure where to go from here.  I've hacked around the
> problem by hard-coding the appropriate strings into the perl script
> and bypassing dbcheck entirely.  But I'd surely like to fix it
> properly...
> 
> Thoughts?
> 
Log a bug on bugs.bacula.org ?

I did it for you see:

http://bugs.bacula.org/view.php?id=1893

(login anonymous/anonymous if you don't have an account)

The fix is rather simple, only dbcheck does a db_init_database and
never an db_open_database (which is also not needed for printing the
database info) but the db_close_database method calls the db_end_transaction
method which does a db_lock which uses the rwl_lock which is still invalide
as it only gets initialized in db_open_database. So the workaround is to
never call db_end_transaction on a non-open database (makes no sense anyhow
as there never will be a transaction when the database is not opened)
and an extra check in the rwl_lock tear down to check if it even is
initialized. Thanks for the analyze however it made my life a lot easier.

I have assigned the bug to me and will be pushing a fix to the next version.

Marco